Prediction related phenomena of visual perception
Perception is grounded in our ability to optimize predictions about upcoming events. Such predictions depend on both the incoming sensory input and on our previously acquired conceptual knowledge. Correctly predicted or expected sensory stimuli induce reduced responses when compared to incorrectly predicted, surprising inputs. Predictions enable an efficient neuronal encoding so that less energy is invested to interpret redundant sensory stimuli. Several different neuronal phenomena are the consequences of predictions, such as repetition suppression (RS) and mismatch negativity (MMN). RS represents the reduced neuronal response to a stimulus upon its repeated presentation. MMN is the electrophysiological response difference between rare and frequent stimuli in an oddball sequence. While both are currently studied extensively, the underlying mechanisms of RS and MMN as well as their relation to predictions remains poorly understood. In the current thesis, four experiments were devised to investigate prediction related phenomena dependent on the repetition probability of stimuli. Two studies deal with the RS phenomenon, while the other two investigate the MMN response. In Experiment 1 the temporal dynamics underlying prediction and RS effects were tested. Participants were presented with expected and surprising stimulus pairs with two different inter-stimulus intervals (0.5s for Immediate and 1.75 or 3.75s for Delayed target presentation). These pairs could either repeat or alternate. Expectations were contingent on face gender and were manipulated with the repetition probability. We found that the prediction effects do not depend on the length of the ISI period, suggesting that Immediate and Delayed cue-target stimulus arrangements create similar expectation effects. In order to elucidate the neuronal mechanisms underlying these prediction effects (i.e. surprise enhancement or expectation suppression), in our second study, we employed the experimental design of the first experiment with the addition of random events as a control. We found that surprising events elicit stronger Blood Oxygen Level Dependent (BOLD) responses than random events, implying that predictions influence the neuronal responses via surprise enhancement. Similarly, the third experiment was employed to disentangle which neural mechanism underlies the visual MMN (vMMN). We compared the responses to the stimuli (chairs, faces, real and false characters) presented in conventional oddball sequences to the same stimuli in control sequences (Kaliukhovich and Vogels, 2014). We found that the neural mechanisms underlying vMMN are category dependent: the vMMN of faces and chairs was due to RS, while the vMMN response of real and false characters was mainly driven by surprise-related changes. So far, no study used category-specific regions of interest (ROIs) to examine the neuroimaging correlates of the vMMN. Therefore, for the fourth experiment, we recorded electrophysiological and neuroimaging data from the same participants with an oddball paradigm for real and false characters. We found a significant correlation between vMMN (CP1 cluster at 400 ms) and functional magnetic resonance imaging adaptation (in the letter form area for real characters), suggesting their strong relationship. Taking the four studies into consideration, it is clear that surprise has an important role in prediction related phenomena. The role of surprise is discussed in the light of these results and other recent developments reported in the literature. Overall, this thesis suggests the unification of RS and MMN within the framework of predictive coding.